Exact computation of queries involve retrieving the full content of all data sets relevant to a predicate, computing the union, applying the predicate to all items in the union, and adding up the weights of items that satisfy the predicate. This procedure is typically too costly or infeasible for massive or distributed data. In some cases, the full data set may no longer be available at the time the query is formulated.